Work Absence in Europe
Author:
Ms. Lusine Lusinyan
Search for other papers by Ms. Lusine Lusinyan in
Current site
Google Scholar Close
and
Mr. Leo Bonato
Search for other papers by Mr. Leo Bonato in
Current site
Google Scholar Close
Publication Date:
30 May 2007
eISBN:
9781589066519
Language:
English
Keywords:
SP; central bank; absence behavior; insurance system; sick pay; insurance scheme; absence rate; Employment protection; Insurance; Labor force participation; insurance provision
Download PDF (481.7 KB)

Work absence is a part of an individual’s decision concerning hours worked. This paper focuses on sickness absence in Europe and builds on an analytical framework in which absence enters both labor supply and demand considerations, with sickness insurance provisions and labor market institutions affecting the costs of absence. The results from a panel of 18 European countries indicate that absence is higher under generous insurance systems and where employers bear little responsibility for their costs. Shorter working hours reduce absence, but flexible working arrangements are preferable if labor supply erosion is a concern.

Abstract

Work absence is a part of an individual’s decision concerning hours worked. This paper focuses on sickness absence in Europe and builds on an analytical framework in which absence enters both labor supply and demand considerations, with sickness insurance provisions and labor market institutions affecting the costs of absence. The results from a panel of 18 European countries indicate that absence is higher under generous insurance systems and where employers bear little responsibility for their costs. Shorter working hours reduce absence, but flexible working arrangements are preferable if labor supply erosion is a concern.

IMF Staff Papers (2007) 54, 475-538. doi:10.1057/palgrave.imfsp.9450016

Low and falling labor utilization has been blamed for the lackluster growth performance of many European countries (OECD, 2003). To a large extent, labor supply erosion can be attributed to the decline in working time. In fact, although participation—possibly owing to labor market reforms—has increased in most European countries in the past 20 years, average working time has continued falling, in line with a long-standing trend, also common to Japan, but not to the United States or Australia (Figure 1).1

Figure 1.
Figure 1.

Labor Force Participation Rate and Average Hours Worked Annually per Employee

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Source: Organization for Economic Cooperation and Development (OECD), OECD Economic Outlook Database.Notes: AU = Australia; BE = Belgium; CA = Canada; CH = Switzerland; DE = Germany; DK = Denmark; ES = Spain; FI = Finland; FR = France; IE = Ireland; IS = Iceland; IT = Italy; JP = Japan; NL = Netherlands; NO = Norway; SE = Sweden; UK = United Kingdom; US = United States.

Declining hours worked can be a reflection of policies as well as changing preferences. Prescott (2004) finds that differences in the marginal tax rate on labor income can explain most of the historical and cross-country variation in labor supply in the Group of Seven countries. Preferences could, however, have also affected the trend of falling working time, which has been a prominent objective of unions in many European countries for some time (Blanchard, 2004). In contrast to Prescott (2004), Alesina, Glaeser, and Sacerdote (2005) argue that European labor market regulations explain most of the difference between Europe and the United States. In any case, this trend presents a challenge for European economies in many ways. With a dwindling labor supply, it is not clear that the current level of potential growth and the financing of large welfare states can be maintained over time. Indeed, the negative consequences for competitiveness have already been triggering pressures to change course in France and Germany.

Actual hours worked may be lower because contractual hours are falling or work absence is rising. In Europe, the decline seems to be driven by the reductions in working time negotiated by unions. In 2003-05, average hours collectively agreed to range from a weekly minimum of 35 hours in France to a maximum of 40 hours in Greece, with most countries having a working week between 37 and 39 hours. The European Union-15 average (together with Norway) has fallen from 38.6 hours in 1999 to 38.0 hours in 2005.2 Looking ahead, the pressure for working-time reductions is likely to continue as unions remain committed to this objective.

If national holidays and annual leave—for which country provisions vary widely—are excluded, absence can be accounted for essentially by sickness. On average, absence due to sickness is not unusually high in Europe. In the period 1995-2003, the share of full-time employees on sick leave was 2.8 percent, on average, which is very close to the 2.6 percent registered in the United States (Figure 2). There are wide differences across countries, however. Absence seems to be particularly high in the Netherlands (6 percent), Sweden (5.2 percent), Norway (5.0 percent), and the United Kingdom (3.9 percent). For these countries, reducing sickness absence could provide a substantial boost to the labor supply.

Figure 2.
Figure 2.

Average Sickness Absence, 1995–2003

(As a percentage of employment)

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Source: Organization for Economic Cooperation and Development (OECD), OECD Economic Outlook Database.Notes: AU = Australia; BE = Belgium; CA = Canada; CH = Switzerland; DE = Germany; DK = Denmark; ES = Spain; FI = Finland; FR = France; IE = Ireland; IS = Iceland; IT = Italy; JP = Japan; NL = Netherlands; NO = Norway; SE = Sweden; UK = United Kingdom; US = United States.

Containing work absence can be beneficial for a number of reasons. Excessive work absence involves significant social and economic costs. In the presence of institutional constraints affecting the choice between work and leisure, such as minimum working hour requirements, absence can be seen as an efficient individual response to the need for flexibility (Dunn and Youngblood, 1986). When absence costs are not internalized by workers, however, significant efficiency costs may arise. Moral hazard may become widespread if insurance is too generous, altering incentives in a way that may not provide the best trade-off between protection and efficiency. Output and employment are likely to be lower in equilibrium owing to the imperfect substitutability of absent workers. If insurance costs are borne mainly by the government, as is the case in most European countries, significant fiscal costs will also arise.3

The main approach in the literature to analyzing labor absence and absenteeism has been based on a standard labor-leisure choice framework (Allen, 1981; and Leigh, 1985). Health, age, gender, and working-time arrangements may influence the preference for leisure. With imperfect monitoring, the decision about sick leave is ultimately left to workers, and moral hazard arises. Its impact can be compounded by changing social norms, a weakening work ethic, and decreasing stigma associated with “benefit cheating” (Lindbeck, 1997). This paper extends the literature to include labor demand considerations, with a role for labor market institutions and sickness insurance systems. Employers’ reaction to absence is likely to depend on the costs they have to bear as a result of it, such as output loss and costs related to insurance schemes (disbursement of cash benefits or contributions to insurance funds). The more costly absence is to employers, the more likely they are to respond. If absence is clearly connected to the working environment, the employer may attempt to improve it. The employer can also increase monitoring or reinforce sanctions for absence. Then labor market institutions come into play. Both employment protection and unemployment insurance reduce the expected cost of work absence to the individual employee either by making it more difficult to sanction absenteeism or reducing the effective cost of the sanction.

The paper contributes to the empirical literature by analyzing the determinants of sickness absence in a panel of 18 European countries during the period 1983-2003, using novel data sets of sickness insurance provisions and costs to employers. The following section describes some key facts about sickness absence in Europe. Section II discusses the model of work absence. Section III elaborates on the econometric issues and presents the results from panel data model estimations. Concluding remarks and policy implications follow in Section IV.

I. The Facts

A glance at the data on sickness absence and some key variables that may affect absence behavior suggests large differences across Europe, in terms of both the importance of the problem and the evolution of the sickness absence and its determinants.

Sickness absence on average is not particularly high in Europe (Figure 2), and the problem seems to be confined to a few countries. Furthermore, Figure 3 shows that sickness absence has been generally stable over the past two decades. There are however some exceptions. In Belgium, for example, the sickness absence rate surged from 1 to 4 percent between 1990 and 2000. Among the countries with the highest absence rates—the Netherlands, the United Kingdom, Sweden, Norway, and Iceland—Sweden has exhibited an upward trend in recent years, whereas the Netherlands has seen absence declining since 1999. Absence in the United Kingdom has remained broadly stable throughout the period. In most countries, sickness absence is higher for women than for men.

Figure 3.
Figure 3.

Sickness Absence

(Employees absent due to sickness as a percentage of total employed)

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Sources: Eurostat, New Cronos Database; and U.S. Department of Labor, Bureau of Labor Statistics.Notes: Bold line is the total sickness absence; the vertical axis for the Netherlands, the United Kingdom, Sweden, Norway, and Iceland has a higher scale.

Sickness absence has often been linked to cyclical fluctuations in the economy. Procyclicality of work absence may arise for two main reasons suggested in the literature (Leigh, 1985; Kaivanto, 1997; and Audas and Goddard, 2001). High unemployment acts as a “disciplining device” (Shapiro and Stiglitz, 1984), raising the expected cost of absence to workers. Others emphasize a “selection” effect, because employers are more likely to lay off absence-prone workers in recessions, and hire them during expansions. Arai and Skogman Thoursie (2001) provide evidence in favor of the market discipline effect in Sweden. However, the strength of procyclicality in countries where employment protection is high may cast some doubt on this interpretation. An alternative explanation could rely on sick leave as a reaction to work pressures, which are likely to be more intense when production volumes are high and labor flexibility is limited. For a selected group of countries, Figure 4 shows how sickness absence changed over time with changes in the cyclical position.4

Figure 4.
Figure 4.

Cyclicality of Sickness Absence

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Sources: Eurostat, New Cronos Database; OECD, OECD Economic Outlook Database; and authors’ calculations.Notes: Vertical axis: employees absent due to sickness as a percentage of total employed; horizontal axis: unemployment gap—percentage deviation of unemployment rate from trend.

The incentives stemming from a country’s insurance system may have a strong impact on absence behavior.5 The sickness insurance systems are most generous in the Nordic countries and Germany (Appendix III; and MISSOC, 2006). Cash benefit replacement rates are high—as high as 100 percent in Norway—with many labor contracts providing for additional benefits from employers. Coverage tends to be universal and benefits are provided for a long period. Sickness benefits can generally be converted into a disability pension if illness continues for a long time. In the past 20 years, most countries have cut replacement rates (Figure 5, left panel). In Finland, for example, the after-tax replacement rate fell by more than 11 percentage points over the past two decades. However, the overall generosity of the system—including also other aspects such as coverage, duration, qualifying and waiting periods—has actually increased in some cases (Figure 5, right panel). In the United Kingdom, for example, the entitlement period has been substantially extended.

Figure 5.
Figure 5.

Changes in Sickness Insurance System, 1983–2002

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Sources: Scruggs (2004); and authors’ calculations.Notes: AT = Austria; BE = Belgium; CH = Switzerland; DE = Germany; DK = Denmark; FI = Finland; FR = France; IE = Ireland; IT = Italy; NL = Netherlands; NO = Norway; SE = Sweden; UK = United Kingdom.

Employers’ responsibility in sharing the costs of the public insurance system can create a stronger incentive for employers to reduce sickness absence. Provisions vary widely across countries. Figure 6 shows a measure of costs to employers of the public insurance system, which reflects the gross replacement rate of benefits paid and their average duration.6 Employers’ costs are highest in Austria and the Netherlands. The Netherlands took a radical approach in 1996, making employers responsible for the full cash benefit payment up to one year of absence. Most firms, however, opted to reinsure their sick pay liability with private insurance companies, reducing the incentive effect. Nonetheless, De Jong and Lindeboom (2004) do not find any difference in absence rates of firms that opted for reinsurance. Although any conclusion from that experience is still tentative, absence started declining three years later and has now dropped below the Swedish level. In general, an analysis of developments since the early 1980s suggests a trend toward shifting more responsibility for sickness insurance costs to employers in most countries.

Figure 6.
Figure 6.

Sickness Cash Benefits Paid by Employer

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Sources: U.S. SSA, Social Security Programs Throughout the World; Eurostat, New Cronos Database; and authors’ calculations.Notes: AT = Austria; BE = Belgium; CH = Switzerland; DE = Germany; DK = Denmark; ES = Spain; FI = Finland; FR = France; GR = Greece; IE = Ireland; IS = Iceland; IT = Italy; LU = Luxembourg; NL = Netherlands; NO = Norway; PT = Portugal; SE = Sweden; UK = United Kingdom.

Finally, the choice of work effort may be influenced by working-time arrangements, with long working hours likely increasing and flexible working arrangements reducing the incidence of absence. Usual (contractual) hours of work show a wide range, with Iceland and the United Kingdom at the top (Figure 7, left panel). The United Kingdom, in particular, presents a large difference between usual hours worked (43.3 hours per week in 2002 and 43.1 hours per week in 2003) and the average working time collectively agreed to between employers and unions (37.2 hours per week in 2002–2003.7 Figure 7 (right panel) indicates that the prevalence of part-time arrangements varies widely between the Netherlands (more than 30 percent of employment) and Greece (about 7 percent).

Figure 7.
Figure 7.

Working-Time Arrangements

Citation: IMF Staff Papers 2007, 001; 10.5089/9781589066519.024.A003

Sources: Eurostat, New Cronos Database; and International Labor Organization (2003).Notes: AT = Austria; BE = Belgium; CH = Switzerland; DE = Germany; DK = Denmark; ES = Spain; FI = Finland; FR = France; GR = Greece; IE = Ireland; IS = Iceland; IT = Italy; LU = Luxembourg; NL = Netherlands; NO = Norway; PT = Portugal; SE = Sweden; UK = United Kingdom.

II. The Model

Theoretical literature on labor absence and absenteeism has focused mostly on the labor supply side.8 This section presents an analysis of work absence within a model that, while still kept simple, combines labor supply and labor demand. Furthermore, the conventional determinants of the labor-leisure choice are augmented in two key areas: (1) a number of institutional characteristics such as the generosity of paid leave provisions and employment protection are introduced; and (2) differences in the impact of publicly and privately financed insurance schemes on work absence are explored.

The economy is populated by a large number of workers whose mass is normalized to 1. A worker’s preference toward absence is given by the desired absence hours, a, such that aє[0, c], where, if contracted hours of work are given by c and total number of hours is normalized to 1, a = 0 indicates no absence, and a = c corresponds to full absence from work.9 Let the worker’s maximization problem be given by

max ⁡   U ( x , l ) , ( 1 )

subject to

x = R + P ( A , V ) w ( c − β a ) + [ 1 − P ( A , V ) ] [ B + G ( γ ) ] ( 2 )
l = 1 − ( c − a ) ⁢ , ( 3 )

where x and l are consumption and leisure, respectively; R is nonlabor income; w is wage; c is the given contractual hours of work; a is absence hours (due to sickness); β is the inverse of the sickness benefit replacement rate (ratio of sick pay to wage) such that βє[0,1], with β = 0 corresponding to the case when sickness absence is fully compensated (100 percent replacement rate) and β = 1 when there is no compensation. B is unemployment benefits. γ is the degree of employment protection and/or level of unionization and can generally be regarded as a combination of labor market regulations that impose costs on employers to discipline or dismiss employees. Assume γє[0,1], where γ = 1 is the situation of “complete” employment protection (no firing possibility) and γ = 0 is the case of no protection at all. G is firing-related entitlements, such that higher entitlements are associated with stronger employment protection, Gγ(γ)>0, and G(0) = 0. The probability of keeping the job (inverse of the penalty for being absent), P(A, V), is a function of the joint impact of absence behavior, a, and employment protection, γ, denoted by A, such that the probability of keeping the job declines with absence, Pa(.)<0, and increases with the degree of employment protection, Pγ(.)>0. P(A, V) is also assumed to depend on the joint impact of some business cycle characteristics, v, and employment protection, γ, denoted by V, such that, if v is a procyclical variable, the probability of remaining employed is higher during upswings Pv(.)>0. Also assume Pav(.) = 0.10

Thus, the budget constraint (Equation 2) states that income spent on consumption is equal to the sum of nonlabor income and wage income if the worker retains her job or unemployment and other firing entitlements if the worker is dismissed.11 In turn, the time constraint (Equation 3) assumes that if total hours are normalized to 1, then total leisure time is the difference between total and actual hours worked.

The firm chooses its desired input of hours of work, which, in the case of a given number of contracted hours, translates into a decision on absence tolerance. It maximizes its profit given by

Π = Y − P ( A ⁢ , V ) w ( c − θ β a ) − [ 1 − P ( A ⁢ , V ) ] G ( γ ) ⁢ , ( 4 )

where

Y = A K 1 − η ( c − a ) η ( 5 )

is the production function, which depends on actual hours worked, (c—a), with the labor share being η. Note that the replacement rate β enters the profit Equation (4) with an additional parameter θ, which indicates whether the insurance system is private or public, that is, whether sick pay is paid by the firm or by the government. In particular, if the employer pays sick pay at the rate β, as set by regulations, then θ = 1. Alternatively, θ = 1/β. We will discuss these two cases separately when deriving the main results below.

From the first-order conditions of the optimization problems in Equations (1)-(3) and (4)-(5) with respect to the absence hours, a, we can obtain the optimal wage for the worker and the firm, respectively, from Equations (6) and (7) below.

w W = P a ( B + G ) − U l U x P a ( c − β a ) − P β ( 6 )
w F = Y a + P a G P a ( c − θ β a ) − P θ β ( 7 )

Case 1: Privately Financed Insurance

When sickness benefits are paid by the firm, θ = 1, equating (6) and (7) yields

U l U x − P a B = − Y a , ( 8 )

stating that in equilibrium, the marginal product of labor (MPL) (-Ya) equals the marginal rate of substitution (MRS) between leisure and consumption, net of the marginal unemployment benefit. Recall that Pa < 0; hence, MPL should be higher or MRS should be lower than in the standard case when PaB = 0.12 Using the Implicit Function Theorem, the following relationship can be obtained for the equilibrium hours of absence, a*:

a * = a ( β − , γ + , c + , v + , B + , R + ) , ( 9 )

whereas the equilibrium wage can be found by evaluating Equations (6) or (7) at the point a = a*. 13

Case 2: Publicly Financed Insurance

Similarly, when the employer does not pay for the sickness absence, and θ = 1/β, from Equations (6) and (7) we have

P a ( B + G ) − U l U x P a ( c − β a ) − P β = Y a + P a G P a ( c − a ) − p , , ( 10 )

which after some algebraic transformations yields

Q + N ( 1 − β ) ( Y a + P a G ) ( M − N ) = 0 , ( 11 )

where Q=Ya+UlUx−PaB, M = PaC, and N = Pa + P. Note that Equation (11) differs from the above case when θ = 1 (Equation 8) by the second term; that is, the difference between equilibrium absence under public and private insurance is given by D=N(1−β)(Ya+PaG)M−N⁢, which will be equal to zero if β = 1 (no compensation for absence) and/or N = 0.14 Rewrite Equation (11) as

U 1 U x − P a B + D = − Y a , ( 12 )

from which it follows that when N>0 (N<0), the wedge between MRS and MPL increases (decreases) and absence in the case of the publicly financed system is higher (lower) than in the case of the employer paying all benefits (D = 0). Observe that the condition N>0 (N<0) can be rewritten as εPa >—1 (εPa—1), where εPaє(— ∞,0] is the elasticity of the probability of keeping the job, P, with respect to absence, a. This implies that a privately financed system will yield lower absence than a publicly financed one if the elasticity of the probability of being fired with respect to absence is low. In other words, if the decision on employment continuation is not very sensitive to absence behavior, then to achieve lower absence it is optimal to shift to employers the responsibility for sickness insurance costs.

The main predictions of the model can be summarized as follows. Sickness absence is expected to decline with the inverse of the replacement rate (or equivalently, increase with the generosity of sickness benefits) and increase with the degree of employment protection and contractual hours. Higher absence is also positively related to cyclical expansions, unemployment benefits, and nonlabor income.15 The direction of the impact of a privately as opposed to publicly financed insurance scheme, given its relationship to various assumptions in the model, largely remains to be determined by the data.

III. Econometric Analysis

Although the empirical literature on work absence in individual countries is vast, there are only a few cross-country comparative studies. Drago and Wooden (1992), using a micro database from 15 plants in the United States, Canada, New Zealand, and Australia, find higher absence rates among women, full-time, low-wage, and long-tenure employees. In their study, absence also appears to be positively correlated with shift work, the generosity of sick leave entitlements, and better labor market options. Using labor force survey data from the Luxembourg Employment Study, Barmby, Ercolani, and Treble (2002) come to similar conclusions. The relationships identified seem to be true for all nine countries in the sample (eight European countries and Canada), in spite of large differences in country mean rates of absence. The authors also identify a robust relationship with hours usually worked, with absence increasing with the number of regular hours worked. Bergendorff and others (2004) investigate the determinants of absence looking at aggregate data from labor force surveys collected by Eurostat for a sample of eight European countries. Showing remarkably high sickness absence in Norway, Sweden, and the Netherlands, their results confirm that sickness absence increases with age, is higher among female employees and, in some countries, is positively correlated with the employment rate, particularly of older workers. They also find some support for the cyclicality of absence, which is shown to be particularly pronounced in Sweden, the Netherlands, and Norway. Moreover, there is evidence that temporary workers, who enjoy lower employment protection, tend to be less sickness prone than permanent workers; however, over time, even though the share of temporary employment has increased, the level of sickness absence in most countries has not declined. Finally, no clear relationships have been found between sickness absence and health status, working conditions, and, largely owing to data limitations, public insurance schemes.

This section discusses the data and empirical results for the determinants of sickness absence in a panel of 18 European countries (Appendix Table A.1). In addition to broadening the country coverage and the range of econometric techniques used previously, the analysis contributes to the existing empirical studies by introducing and exploring novel data sets on sickness insurance provisions and costs to employers as well as controlling for labor market regulations.

Data

The data on sickness absence draw on labor force surveys and, particularly, on the Eurostat Labour Force Survey Results, which include aggregated data on average usual and actual hours of work. Our definition of absence includes both short-term (at least one hour) and long-term (at least one week) absences,16 unlike Bergendorff and others (2004), who use only long-term absence data. Data on age, health, unemployment, and participation are drawn from the International Labor Organization’s Key Indicators of the Labour Markets (ILO, 2003). Data on institutional characteristics of social security systems are derived from Scruggs (2004). Data on the cost to employers of the sickness insurance system are based on information from the U.S. Social Security Administration and social security programs throughout the world (Appendix Table A.2). Basic descriptive statistics of the variables used in the analysis and their cross-correlations are summarized in Appendix Table A.3 and A.4, respectively.

Empirical Strategy

The econometric exercise is based on standard panel data models, with particular consideration given to their performance in the context of macroeconomic applications. Though extensive cross-sectional information, typical of microeconomic data sets, is not available in such cases, the panel data approach allows us to analyze sickness absence developments over time and across countries. The availability of working-time data and some of the absence determinants by gender makes it possible to combine sickness absence for males and females and double the effective cross-sectional dimension of the panel data.

In a general setup, the model is given by

a i , t = Σ j = 1 k a i , t − j β 0 , j + X i , j β 1 + W i , j β 2 + η i + ɛ i , t , i = 1 , … N ; t = 1 , … , T i , ( 13 )

where ai,t is the absence rate for country-gender pair i at time t, Xi,t is a vector of exogenous covariates, and Wi,t is a vector of predetermined and endogenous covariates treated similarly to the lagged dependent variables. X and W may contain lagged independent variables and time dummies, and can be either country-gender or only country specific. ηi is an unobserved unit-specific fixed effect, and εi,t is the disturbance term.

The determinants of sickness absence have been estimated with both static (β0, j = 0) and dynamic (β0, j ≠ 0) panel data (DPD) models that control for fixed effects. The former follows the common approach applied in the earlier literature; the latter, however, allows us to build richer dynamics into the relationship between absence and its determinants by taking into account potential persistence in absence rates and endogeneity of some right-hand-side variables. Furthermore, DPD models can help us address the error autocorrelation problem, which, if ignored, results in consistent but inefficient estimates of regression coefficients and biases the inference. Serial correlation, in turn, may occur, mostly as a result of omitting variables that change gradually over time. Because the inclusion of a lagged dependent variable makes the least squares dummy variable (LSDV) and generalized least squares (GLS) estimators biased and inconsistent for finite T, an instrumental variable or a generalized method of moments (GMM) estimation method can be used. Monte Carlo results of Judson and Owen (1999) suggest that, in macroeconomic panel data applications, the one-step GMM estimator by Arellano and Bond (1991) is a second-best choice.17 Furthermore, as shown in Blundell and Bond (1998), persistence in the dependent variable may result in weak instruments and losses in asymptotic efficiency when using the first-difference GMM estimator. As an alternative, the system GMM estimator is suggested, which combines the regressions in differences used in the standard first-difference GMM estimation with the regressions in levels.

Our empirical strategy has been the following. For each specification, static (fixed-effects, random-effects, and pooled ordinary least squares (OLS)) and dynamic (GMM; Anderson and Hsiao, 1982) panel data models have been estimated.18 For static models, the appropriateness of the random-effects specification has been tested by the Hausman test, together with a test of serial correlation in idiosyncratic error terms (Wooldridge, 2002; and Drukker, 2003). In a dynamic setup, we have used a one-step GMM estimator, which generally tends to be less biased in small samples than the two-step estimator and outperforms the latter in macroeconomic applications (Judson and Owen, 1999; and Lusinyan, 2005). The two-step estimator has, however, been used for robustness checks, along with the LSDV estimator with the Kiviet’;s (1995) correction. The one-step estimations have been implemented using first-difference and system GMM estimators, with standard errors assumed to be both robust and nonrobust to general heteroscedasticity over individuals and over time. The results have also been checked by using different sets of GMM instruments. Although the GMM procedure of Arellano and Bond (1991) implies using all lagged values as instruments, Judson and Owen (1999) argue that a “restricted GMM,” with a reduced number of values of the lagged dependent variables and exogenous regressors used as instruments, does not substantially affect the performance of this technique.19 Finally, robustness of the results to the exclusion of some countries from the sample as well as restricting the dependent variable to short- or long-term sickness absences have been checked.

Before presenting the main results, it is worth discussing endogeneity. In a static setup, some covariates, such as the demographic variables (labor force participation rate, age structure of the labor force, and health status) can be assumed as given by construction for the sickness absence behavior in a certain year.20 When we have a choice between alternative measures for the same determinant, such as the share of part-time employment and absence because of a flexible working-time arrangement, we do not use the variable, which is likely to be endogenous by construction (for example, in this case, the share of people absent because of a flexible working-time arrangement). The section on the robustness analysis discusses the findings from exogeneity tests, particularly for the usual hours of work and sickness insurance characteristics. With little evidence to support the endogeneity of these covariates by the data, addressing error autocorrelation and potential omitted variable bias remains a key consideration in the static setup.

In the dynamic models, however, the assumptions of exogeneity can be challenged. For example, the current level of sickness absence may affect future realizations of demographic variables, especially if current sickness is correlated with disability and exit from the labor force, as discussed earlier. Similarly, it can be suggested that taking sick leave when needed would positively influence an employee’s health, allowing more people to remain longer in the labor force and possibly also increasing life expectancy. We can also expect some intertemporal relationship between the sickness absence rate among full-time employees in a given year (our dependent variable) and the share of part-time employees, because health conditions may require employees to quit full-time employment and work part-time. To address the potential endogeneity problem, most of these covariates have been modeled as predetermined variables in the dynamic specifications, with the Sargan test for overidentifying restrictions supporting this choice.

Discussion of Results

The results from the static and dynamic panel data regressions are reported in Tables 1 and 2, respectively. Lower labor force participation and better health (here, longer life expectancy) are shown to be associated with lower sickness absence.21 The positive relationship between labor force participation and sickness absence seems to hold with a lag in the dynamic models, with the contemporaneous effect being negative but mostly insignificant. In addition, even though the contemporaneous effect of life expectancy remains negative in the dynamic models, the overall intertemporal impact could be positive if dominated by a strong positive lagged effect. A larger share of workers aged 55-64 in the total labor force increases sickness absence, but this result is less robust.

Table 1.

Determinants of Sickness Absence: Static Panel Data Models

(Dependent variable: share of employees absent due to sickness in total employed)

article image
Notes: The table reports the results from the fixed-effects models with robust errors. Annual panel data over the period 1983-2003 (time periods vary, see Appendix Table A.1) are used including the following groups of countries: 18 countries in columns 1-4 (see Table A.1); 13 countries in columns 5-7 (countries in Table A.1, excluding Greece, Spain, Luxembourg, Portugal, and Iceland); and 15 countries in column 8 (countries in Table A.1, excluding Greece, Luxembourg, and Iceland). The cross-sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. Robust t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment. See Appendix Table A.2 for the definitions and sources of the variables.
Table 2.

Determinants of Sickness Absence: Dynamic Panel Data Models

(Dependent variable: share of employees absent due to sickness in total employed)

article image
article image
Notes: The table reports the results from the Arellano-Bond (1991) one-step GMM models with restricted set of instruments and robust standard errors. Annual panel data over the period 1983-2003 (time periods vary, see Appendix Table A.1) are used including the following groups of countries: 18 countries in columns 1-4 (see Table A.1); 13 countries in columns 5-7 (countries in Table A.1, excluding Greece, Spain, Luxembourg, Portugal, and Iceland); and 15 countries in column 8 (countries in Table A.1, excluding Greece, Luxembourg, and Iceland). The cross-sectional unit, i, is the country- gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. AR(2) is the test of the null of no-second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions. Robust t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment. See Appendix Table A.2 for the definitions and sources of the variables.

The results from both static and dynamic models show the existence of a significant impact of working-time arrangements on sickness absence. In particular, usual hours worked are estimated to have a strong positive impact on sickness absence, whereas more flexibility—measured by the share of part-time employment and flexible working-time arrangements—helps reduce sickness absence, both directly and through interaction terms. The estimated impact of usual hours worked appears to be close to the findings by Barmby, Ercolani, and Treble (2004) for the United Kingdom, for which the estimated coefficient of usual hours is 0.16. Indeed, a major conclusion of Barmby, Ercolani, and Treble is that sickness absence is relatively more sensitive to the determinants that measure contractual arrangements than to individual characteristics. Our results suggest that an increase in the average usual hours by one hour will increase the average absence rate from 2.75 to 2.9 percent, whereas with just a 1 percentage point increase in the average share of part-time employment, the absence rate will decline to 2.45 percent. Among the determinants considered in this paper and based on the assumptions about their possible changes, the working-time arrangements appear to have the most economically significant impact on absence behavior. In discussing the robustness of the results, we show that the relationship between the working- time arrangements and sickness absence is particularly strong for short-term sickness absence.

The unemployment gap is estimated to have a negative sign, implying that an increase in the gap between the unemployment rate and its trend—a proxy for a cyclical contraction—would reduce sickness absence, in line with the hypothesis that market conditions exert a disciplining effect on absence. It appears, however, that the size and significance of the impact of the unemployment gap are to some extent driven by the Swedish data, as discussed in the section on the robustness analysis.

As expected, we find a positive relationship between sickness absence and the generosity of the sickness insurance scheme or sickness benefits, as measured by the after-tax replacement rate. The robustness checks further show that, in contrast to working-time arrangements, the effect of sickness benefits is stronger on long-term absence. Unemployment benefits are shown to have a negative impact on sickness absence, which appears to be at variance with the predictions of the model. However, it should not be surprising, given that our dependent variable measures the share of full-time employees absent due to sickness. Indeed, even though in the model a higher unemployment benefit implies higher income in the event a worker is fired because of absence—and thus weakens the disciplining effect—for the data used, this would translate into a lower number of full-time employees (total and on sick leave), which will result in a smaller share of absentees.

There is some evidence in the data that absence declines when employers bear larger costs of sickness insurance. Measured as the product of the cash benefit replacement rate and the period that falls under the employer’s responsibility, the employer-sick-pay variable could have both a positive (increase in generosity of sick pay) and negative (stricter monitoring) impact on absence. Our findings suggest that the employer’s incentive-and-behavior effect generally outweighs that of employees, particularly in the case of long-term absence.22

Finally, labor market institutions affect the absence rate both directly and through their interaction with sickness insurance provisions and business cycle characteristics. Given the assumptions of the model, we expect to find a positive relationship between the degree of employment protection and sickness absence. This relationship appears strongest for the level of unionization, implying, for example, that a 1 percentage point reduction in average union density (with an elasticity at the mean estimated at about 0.6) will lower the average absence rate from 2.75 to 2.71 percent.23 Furthermore, the data support the hypothesis that the negative impact of an employer’s sick pay provision appears to decrease somewhat with the degree of unionization (as shown by the positive interaction term), suggesting that the latter may reduce the employer’s ability to enforce better work attendance. The robustness analysis shows that this result is particularly important for long-term sickness absence.

Robustness Analysis

This section discusses the robustness of the results presented in Tables 1 and 2, based on a number of alternative specifications and tests reported in Appendix II.

The results concerning the relationship between sickness absence and demographic characteristics remain largely robust across various specifications (Appendix Table A.5), including models with robust standard errors adjusted for within-group (here, country-gender pair) correlation (columns 1 and 3);24 random-effects models, supported by the Hausman and Breusch-Pagan LM tests (not reported here) (columns 2 and 3); pooled OLS with country dummies (column 4); the Anderson-Hsiao (1982) estimator (column 6); as well as Arellano-Bond one-step GMM, when the dependent variable is long-term and short-term sickness absence (columns 7 and 8, respectively). The significant coefficient for the gender dummy indicates that women are more likely to be on sick leave than men; the large and highly significant coefficient estimates on the country dummies for the Netherlands, Sweden, Norway, and the United Kingdom once again support the facts concerning relatively high sickness absence rates in these countries (Section I). The health variable has a similar impact on both long- and short-term sickness absence, but other demographic variables seem to be more important for long-term sickness absence in terms of the persistence and size of their impact.

Moving to the robustness tests for the relationship between sickness absence and working-time arrangements, we first discuss the issue of possible endogeneity of this set of determinants. Although the decision to be absent is made given specific working-time arrangements, absence behavior may have implications for the employer’s reaction. Assuming that employers want to maintain the same total hours of work and employment regulations permit such adjustments, employers may opt to demand longer hours of work from employees other than the ones prone to sickness absence. In line with the model’s predictions, institutional characteristics of the labor market, such as the degree of employment protection, strength of trade unions, and employees’ bargaining power would be important in determining the relationship between sickness behavior and employers’ reaction, including the probability of imposing a penalty for being absent.25 In a specification (not reported here) with labor force participation and usual hours of work, instrumenting the latter with the employment protection index and union density is strongly supported by the Hansen- Sargan test of instrument validity and C-statistic test of exogeneity of instruments.26 At the same time, the Davidson-MacKinnon (1993) test for exogeneity for a fixed-effects regression does not reject the consistency of OLS estimates.

Appendix Table A.6 reports some robustness results that confirm the significance of working-time arrangements for sickness absence behavior. Columns 1–3 are variations of the specifications reported in Table 1 for random-effects models and with robust standard errors adjusted for within- group correlation, and columns 5-7 report some further specifications for dynamic models with different combinations of covariates. Sickness absence is again shown to increase with the usual hours of work, but the impact of the latter decreases with the availability of more flexible working-time arrangements (that is, a higher share of part-time employment or a higher absence rate owing to flexible working-time arrangements). Alternatively, more flexible work arrangements are associated with lower sickness absence, but this relationship weakens with an increase in usual working hours. As is also shown in Table 2, the results for the dynamic models support the importance of the lagged impact of the share of part-time employment (column 6).

The impact of working-time arrangements is particularly significant and robust for short-term absence (Appendix Table A.6, columns 4 and 8). In the case of long-term (and hence, total) absence, the results for flexible work arrangements appear to be sensitive to the exclusion of Belgium (especially, Belgian women), where the long-term absence rate increased sharply in 1998 (Figure 3), likely reflecting the extension of various sabbatical leave programs. The results (not reported here), however, remain robust to the exclusion of the countries with the longest usual hours, Iceland and the United Kingdom.27

Appendix Table A.7 shows several extensions of the previous regressions that control for cyclical developments in the economies. The negative relationship between sickness absence and the unemployment gap continues to hold for alternative definitions of the latter, although the use of a quadratic trend in calculating the unemployment gap improves the significance of its estimated coefficient. The results are, however, somewhat sensitive to excluding the Swedish data, in which case the unemployment gap becomes less significant (compare column 3 in Appendix Table A.7 with column 4 in Table 1, and column 7 in Appendix Table A.7 with column 4 in Table 2).

The robustness checks for the results concerning the relationship between sickness absence and various forms of insurance schemes (sickness insurance, unemployment insurance, and the sick pay financed by employers) are summarized in Appendix Tables A.8 and A.9.28 Similarly to the above discussion concerning potential endogeneity of usual hours of work demanded by the employer, one could argue that sickness insurance characteristics, such as the benefit replacement rate and the costs paid by the employer, may react to higher sickness absence. We analyze possible endogeneity of these regressors using as instruments a number of labor market institutional variables such as the employment protection index, union density, and wage bargaining coordination, and find strong support (not reported here) for non-IV fixed-effects specifications.

The positive impact of sickness benefits on absence is shown to hold in various specifications, but in the static model, the statistical significance of these estimates appears to be sensitive to the exclusion of Sweden from the sample, as well as affected by some collinearity with the life expectancy variable.29 In contrast to working-time arrangements, sickness benefits are more significant for long-term absence (Appendix Table A.8, column 3 and Appendix Table A.9, columns 2 and 6), and are robust to changes in the country sample. Generally, dynamic models provide more robust estimates of the impact of sickness benefits.

The results are less conclusive when the index of sickness insurance system generosity is included instead of or together with the sickness benefits variable. First, in static panel data models, the impact of this variable is insignificant for total sickness absence (Appendix Table A.8, column 4); although it has the expected positive sign for short-term absence, the estimate is not robust to the exclusion of the United Kingdom from the sample (not reported here). Second, for long-term absence, it shows a significant negative influence (not reported here), but, by construction, there might be important lagged effects because the index includes the share of the labor force with sick pay insurance, and, as discussed earlier, some intertemporal relationship between long-term sickness absence and labor force participation could exist. Indeed, in the dynamic models, the specification tests suggest that the sickness index is better specified as a predetermined variable and has a relatively significant positive lagged effect on sickness absence (Appendix Table A.9, column 4). However, the overall robustness of the results with the sickness index is low.

The relationship between the generosity of the unemployment insurance system and sickness absence remains negative, but with a lagged impact in the dynamic models. The results are especially robust when instead of the unemployment benefit variable the index of unemployment insurance system generosity (Appendix Table A.8, column 5, and Appendix Table A.9, column 5) or the combined index of sickness and unemployment insurance systems generosity, without a lag, is included (not reported here).

The robustness checks also confirm that absence declines when employers bear a larger cost for sickness insurance (Appendix Table A.8, columns 6-7), though the results are somewhat sensitive to excluding Belgium. The employer sick pay appears to be particularly strongly (and negatively) correlated with long-term absence (Appendix Table A.9, column 6). Importantly, as noted in the previous section, when the data are partitioned into two subsamples—more vs. less stringent employment protection or a higher vs. lower level of unionization, choosing as a cutoff point the sample mean or median—the coefficient estimate of the employer sick pay variable has a more significant negative value in the subsample corresponding to stricter labor regulations (Appendix Table A.8, column 8) compared with the one with lower employment protection (Appendix Table A.8, column 9) or weaker trade unions.30 These findings lend support to the results from our model concerning the factors that could determine the direction of the impact of privately as opposed to publicly financed sickness insurance on sickness absence.

Finally, the role of labor market institutions is analyzed more explicitly in the last set of regressions, reported in Appendix Table A.10. We investigate both direct and indirect impacts of institutional indicators on sickness absence, including the index measuring the strictness of employment protection, the density of union membership, and the degree of wage bargaining coordination.31,32 The direct impact of these variables on absence behavior may, however, be ambiguous. For example, on the one hand, greater employment protection such as large firing costs may lower the probability of being fired and hence weaken work discipline and increase absence incidents.33 On the other hand, if greater employment protection leads to higher average unemployment, as argued in the literature, the disciplining impact of unemployment would increase the expected cost of absence to workers and result in lower absence. Similar arguments could be put forward when considering the impact of greater unionization. The ambiguity of the impact of wage bargaining coordination can be even greater, because it can have opposite effects on wages and unemployment (IMF, 2004), making it difficult to draw implications for absence behavior. In view of the importance of such interactions, the regressions reported in Tables 1-2 and Appendix Table A.10 look at the impact of some of the above determinants of sickness absence when they are interacted with the indicators of labor market institutions.

The robustness tests seem to reflect the ambiguity concerning the direct impact of labor market institutions on sickness absence. In particular, although the direct impact of the employment protection index on sickness absence is negative (Appendix Table A.10, columns 1 and 5), it is statistically insignificant in a country sample that excludes Belgium. The positive impact of union density, instead, is shown to affect sickness absence (column 2) positively, but is similarly sensitive to changes in the country sample. The impact of wage bargaining coordination is found to be negative and significant (not reported here), but it is not robust to using an alternative version of the index capturing more nuanced variations in the institutional structure. However, the results are more robust when an indirect impact of the institutional variables is considered. Most notably, the results lend further support to the hypothesis that a higher degree of unionization may reduce the negative impact of an employer’s sick pay provision. This result seems to be particularly significant for long-term sickness absence (Appendix

Table A.10, columns 4 and 8); for short-term absence, the estimates are less robust, particularly to changes in the country sample.

IV. Conclusions and Policy Implications

The evidence presented in this paper suggests that sickness absence is very high at least in four countries: the Netherlands, Sweden, Norway, and the United Kingdom. In these countries, between 4 and 6 percent of employees are absent on a given day, with losses in terms of forgone output that are likely to be substantial. Owing to their generous public insurance systems, the Netherlands, Sweden, and Norway bear significant costs in terms of public finances. Containing sickness absence would help prevent erosion of the labor supply stemming from demographics and working-time reductions.

High sickness absence reflects, to some extent, high labor force participation, particularly of women and older people. Countries with high sickness absence have generally high participation rates, to which both the traditional Nordic emphasis on social inclusion and the market-oriented approach followed by the United Kingdom may have contributed. Going forward, as populations age, maintaining high employment rates will be increasingly challenging and containing the erosion of the labor supply even more urgent. With large changes in the composition of the labor force, the overall impact of these changes on sickness absence is difficult to predict.

The high level of sickness absence, however, is not a necessary price for high participation. The results presented in this paper, as well as the evidence provided by the literature, point to a significant incentive problem owing to the generosity and leniency of public insurance schemes, especially in the Netherlands, Sweden, and Norway. Streamlining such systems could help improve labor supply incentives, with the benefits of a well-designed reform likely to be substantial (of the order of 0.5 to 1 percent of GDP for Sweden; Andersen and Molander, 2003). A comprehensive reform of sickness insurance should also consider how it is linked to other components of the social insurance system.34

A shift of a portion of insurance costs to employers may be advisable. The experience of the Netherlands, where absence has declined following a major reform in 1996 that shifted all costs to employers, could provide important lessons. This paper shows that higher costs are likely to produce a response by employers, which would ultimately help reduce absence. This effect, however, is likely to be smaller, the higher the level of employment protection. To be most effective, the cost shift must affect employer incentives via an increase in the marginal cost of absence. If the incentive is diluted, and the shift translates into a mere increase in labor costs, negative effects on employment are more likely to result. One way to achieve a more efficient impact would be to leave more room for workers and employers to determine the level of protection; for example, by a substantial reduction in the replacement rate of the public insurance system.

Encouraging flexible work arrangements is likely to pay off. The results presented here suggest that policies promoting shorter working hours may not be inconsistent with the objective of reducing absence. High sickness absence in the United Kingdom, for example, seems to be explained mainly by its comparatively long working hours. These policies, however, may still lead to a net reduction in hours worked, even if the accompanying decrease in sickness absence would partly offset their effect. Promoting flexible work arrangements, which have been shown to substantially reduce absence, may be a better policy option.

APPENDIX I

Data and Descriptive Statistics

See Tables A.1-A.4.

Table A.1.

List of Countries and Data Availability

article image
Notes: *denotes missing data for Germany (1984), Italy (1992), and the Netherlands (1984, 1986). Working-time data are from the Eurostat New Cronos Database (See Appendix Table A.2).
Table A.2.

List of Variables, Definitions, and Sources

article image
article image
Table A.3.

Descriptive Statistics

article image
Note: See Appendix Table A.2 for the definitions and sources of the variables.
Table A.4.

Cross-Correlations Between Variables of Model

article image
Notes: * = significant at the 5 percent level; LFPR = labor force participation rate; UE = unemployment; EP = employment protection. See Appendix Table A.2 for the definitions and sources of the variables.

APPENDIX II

Robustness Analysis

See Tables A.5-A.10.

Table A.5.

Determinants of Sickness Absence: Demographic Characteristics

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
Notes: Annual panel data over the period 1983-2003 (time periods vary) for 18 countries are used (see Appendix Table A.1). Reported specifications include (1) fixed-effects (FE) model with robust errors adjusted for correlation within country-gender groups; (2) random-effects (RE) model with robust errors; (3) RE model with robust errors adjusted for correlation within country-gender groups; (4) pooled OLS with robust errors; country dummies for other countries are not reported; (5) Arellano-Bond (1991) one-step GMM (AB) model with restricted set of instruments and non-robust errors; (6) Anderson-Hsiao (1982) two-stage least-squares first-differenced estimator; (7) dependent variable is long-term sickness absence; AB model with restricted set of instruments (similar results with robust errors) and three lags of the dependent variable (only the first lag is reported); (8) dependent variable is short-term sickness absence; AB model with restricted set of instruments (similar results with robust errors). The cross-sectional unit, i, is the country-gender pair. All regressions include time-fixed effects (not reported) that are statistically significant at least at the 10 percent level (particularly for the years 1993 and 1994). AR(2) is the test of the null of no-second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions. t-values are in parentheses. **(*,+,-) = significant at 1 (5, 10, 15) percent level. LFPR = labor force participation rate. See Appendix Table A.2 for the definitions and sources of the variables.
Table A.6.

Determinants of Sickness Absence: Working-Time Arrangements

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
article image
Notes: Annual panel data over the period 1983-2003 (time periods vary) for 18 countries are used (see Appendix Table A.1). Reported specifications include (1) random-effects (RE) model with robust errors; similar results for models excluding Iceland and the United Kingdom; (2) RE model with robust errors; (3) fixed-effects (FE) model with robust errors adjusted for correlation within country-gender groups (higher significance with non-cluster robust errors); in RE model with robust errors, part-time employment and its interaction with usual hours of work are significant when Iceland is excluded; (4) dependent variable is short-term sickness absence; FE model with robust errors adjusted for correlation within country-gender groups (higher significance with non-cluster robust errors); similar results for RE model; (5)-(8) Arellano-Bond (1991) one-step GMM model with restricted set of instruments and robust errors; (8) dependent variable is short-term sickness absence. The cross-sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. AR(2) is the test of the null of no- second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions. t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate. See Appendix Table A.2 for the definitions and sources of the variables.
Table A.7.

Determinants of Sickness Absence: Cyclically and Unemployment Gap

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
article image
Notes: Annual panel data over the period of 1983-2003 (time periods vary) for 18 countries are used (see Appendix Table A.1), except in column 3 (see below). Reported specifications include (1) random-effects (RE) model with robust errors; similar results for fixed-effects (FE) model; (2)-(3) FE model with robust errors; (3) excluding Sweden; (4) FE model with robust errors adjusted for correlation within country-gender groups (higher significance with non-cluster robust errors); (5)-(7) Arellano-Bond (1991) one-step GMM model with restricted set of instruments and robust errors; unemployment gap is modeled as predetermined, as supported by the test of instrument validity; (7) excluding Sweden. The cross-sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. AR(2) is the test of the null of no-second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions, t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment. See Appendix Table A.2 for the definitions and sources of the variables.
Table A.8.

Determinants of Sickness Absence: Sickness Insurance (1)

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
article image
Notes: Annual panel data over the period 1983-2003 (time periods vary, see Appendix Table A.1) for 13 countries (see Figure 5) are used, except in columns 8-9 (see below). Reported specifications include (1)-(9) fixed-effects models with robust errors adjusted for correlation within country-gender groups (higher significance with non-cluster robust errors); similar results for random-effects (RE) models with high overall R2 (the latter is in the range of 0.20-0.30 if regressions are modeled as RE models); (3) dependent variable is long-term sickness absence. Specifications in (8)-(9) correspond to subsamples where the employment protection (EP) variable is used to partition the data set such that EP > 1.25 and EP < 1.25 in (8) and (9), respectively; the value 1.25 is the sample mean for EP. Countries in (8) are Belgium, Germany, Spain, France, Italy, the Netherlands, Austria, Portugal, Sweden, and Norway; countries in (9) are Belgium, Denmark, Ireland, the Netherlands, Finland, the United Kingdom, and Switzerland. The results are robust to other choices of the cutoff value, such as 1.32 (sample median) as well as to considering short- and long-term absence separately; similar results are obtained when the partitioning is done using a union density variable, with cutoff values at 45, 42 (sample mean), or 50 (average value of the variable). The cross- sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level, t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment. See Appendix Table A.2 for the definitions and sources of the variables.
Table A.9.

Determinants of Sickness Absence: Sickness Insurance (2)

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
article image
Notes: Annual panel data over the period 1983-2003 (time periods vary, see Appendix Table A.1) for 13 countries (countries in Table A.1, excluding Greece, Spain, Luxembourg, Portugal, and Iceland) are used. Reported specifications include Arellano-Bond (1991) one-step GMM (AB) model with a restricted set of instruments and robust standard errors (similar results with robust errors), with Sargan test from the specification with non-robust errors; (1) sickness benefit is modeled as exogenous, as supported by the test of instrument validity; (2) dependent variable is long-term sickness absence; (4) the sickness index is modeled as predetermined, as supported by the test of instrument validity; (5) the index of unemployment insurance system generosity is modeled as exogenous, but there are similar results if it is modeled as predetermined; (6) dependent variable is long-term sickness absence; employer sick pay is modeled as exogenous, but there are similar results if it is modeled as predetermined; (7) dependent variable is short-term sickness absence. The cross-sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. AR(2) is the test of the null of no-second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions, t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment. See Appendix Table A.02 for the definitions and sources of the variables.
Table A.10.

Determinants of Sickness Absence: Labor Market Institutions

(Dependent variable: share of employees absent due to sickness in total employed, unless otherwise indicated)

article image
article image
Note: Annual panel data over the period 1983-2003 (time periods vary, see Appendix Table A.1) are used including the following groups of countries: 13 countries in columns 1-2 and 5-6 (countries in Table A.1, excluding Greece, Spain, Luxembourg, Portugal, and Iceland); and 15 countries in columns 3-4 and 7-8 (countries in Table A.1, excluding Greece, Luxembourg, and Iceland). Reported specifications include (l)-(2) fixed-effects (FE) model with robust errors adjusted for correlation within country-gender groups (higher significance with non-cluster robust errors); similar results for random-effects (RE) model; (3)-(4) FE models with robust errors (lower significance with cluster robust errors but higher significance in RE model); (4) dependent variable is the long-term sickness absence; (5)-(8) Arellano-Bond (1991) one-step GMM (AB) model with restricted set of instruments and robust standard errors; labor market institutional variables are modeled as exogenous; there are similar results when the sickness benefit variable is dropped, increasing the number of groups to 30, as in (7) and (8); (8) dependent variable is the long-term sickness absence. The cross-sectional unit, i, is the country-gender pair. All regressions include time fixed effects (not reported) that are statistically significant at least at the 10 percent level. AR(2) is the test of the null of no-second-order autocorrelation in the first-differenced residuals, and the validity of instruments is tested using the Sargan test of overidentifying restrictions. t-values are in parentheses. **(*,+,-) = significant at the 1 (5, 10, 15) percent level. LFPR = labor force participation rate, UE = unemployment, EP = employment protection. See Appendix Table A.2 for the definitions and sources of the variables.

APPENDIX III

Sickness Benefits in Europe

See Table A.11

Table A.11.

Comparative Table on Sickness Cash Benefits

(As of January 1, 2006)

article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
article image
Source: MISSOC (2006).

REFERENCES

  • Alesina, A., E. Glaeser, and B. Sacerdote, 2005, “Work and Leisure in the U.S. and Europe: Why So Different?,” NBER Working Paper No. 11278 (Cambridge, Massachusetts, National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Allen, S.G., 1981, “An Empirical Model of Work Attendance,” Review of Economics and Statistics, Vol. 63 (February), pp. 77–87.

  • Andersen, T., and P. Molander, 2003, “Policy Options for Reforming the Welfare State,” in Alternatives for Welfare Policy, ed. by T. Andersen and P. Molander (Cambridge, United Kingdom, Cambridge University Press).

    • Search Google Scholar
    • Export Citation

  • Anderson, T.W., and C. Hsiao, 1982, “Formulation and Estimation of Dynamic Models Using Panel Data,” Journal of Econometrics, Vol. 18 (January), pp. 47–82.

    • Search Google Scholar
    • Export Citation

  • Andrén, D., 2001a, “Long-Term Absenteeism Due to Sickness: The Swedish Experience, 1986-1991,” Working Paper in Economics No. 47 (Göteborg, Sweden, Department of Economics, Göteborg University). Available via the Internet: http://www.handels.gu.se/epc/data/html/pages/PDF/gunwpe0047.pdf.

    • Search Google Scholar
    • Export Citation

  • Andrén, D., 2001b, “Short-Term Absenteeism Due to Sickness: The Swedish Experience, 1986-1991,” Working Paper in Economics No. 46 (Göteborg, Sweden, Department of Economics, Göteborg University). Available via the Internet: http://www.handels.gu.se/epc/data/html/pages/PDF/gunwpe0046.pdf.

    • Search Google Scholar
    • Export Citation

  • Andrén, D., 2003, “Sickness-Related Absenteeism and Economic Incentives in Sweden: A History of Reforms,” CESifo DICE Report 3/2003 (Munich, Center for Economic Studies, Ifo Institute for Economic Research).

    • Search Google Scholar
    • Export Citation

  • Arai, M., and P.Skogman Thoursie, 2001, “Incentives and Selection in Cyclical Absenteeism,” FIEF Working Paper No. 167 (Stockholm, Trade Union Institute for Economic Research).

    • Search Google Scholar
    • Export Citation

  • Arellano, M., and S. Bond, 1991, “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations,” Review of Economic Studies, Vol. 58 (April), pp. 277–97.

    • Search Google Scholar
    • Export Citation

  • Audas, R., and J. Goddard, 2001, “Absenteeism, Seasonality, and the Business Cycle,” Journal of Economics and Business, Vol. 53 (July), pp. 405–19.

    • Search Google Scholar
    • Export Citation

  • Barmby, T.A., M.G. Ercolani, and J.G Treble, 2002, “Sickness Absence: An International Comparison,” Economic Journal, Vol. 112 (June), pp. F315–F31.

    • Search Google Scholar
    • Export Citation

  • Barmby, T.A., M.G. Ercolani, and J.G Treble, 2004, “Sickness Absence in the UK: 1984-2002,” Swedish Economic Policy Review, Vol. 11, pp. 65–88.

    • Search Google Scholar
    • Export Citation

  • Bergendorff, S., and others, 2004, “Sickness Absence in Europe—A Comparative Study,” Social Insurance Studies No. 2 (Swedish National Insurance Board).

    • Search Google Scholar
    • Export Citation

  • Blanchard, O., 2004, “The Economic Future of Europe,” NBER Working Paper No. 10310 (Cambridge, Massachusetts, National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Blundell, R., and S. Bond, 1998, “Initial Conditions and Moment Restrictions in Dynamic Panel Data Models,” Journal of Econometrics, Vol. 87 (August), pp. 115–43.

    • Search Google Scholar
    • Export Citation

  • Brown, S., and J.G. Sessions, 1996, “The Economics of Absence: Theory and Evidence,” Journal of Economic Surveys, Vol. 10 (March), pp. 23–53.

    • Search Google Scholar
    • Export Citation

  • Bruno, G.S.F., 2004, “Approximating the Bias of the LSDV Estimator for Dynamic Unbalanced Panel Data Models,” Studie quaderni dell’Istituto di Economia Politica EEA 2004-1 (Milan, Universita Bocconi) (presented at 10th UK Stata Users Group Meetings, June 2004). Available via the Internet: http://www.stata.com/meeting/10uk/WP2004-1.pdf.

    • Search Google Scholar
    • Export Citation

  • Bun, M.J.G., and J.F. Kiviet, 2006, “The Effects of Dynamic Feedbacks on LS and MM Estimator Accuracy in Panel Data Models,” Journal of Econometrics, Vol. 132 (June), pp. 409–44.

    • Search Google Scholar
    • Export Citation

  • Davidson, R., and J. MacKinnon, 1993, Estimation and Inference in Econometrics (New York, Oxford University Press).

  • De Jong, P., and M. Lindeboom, 2004, “Privatisation of Sickness Insurance: Evidence from the Netherlands,” Swedish Economic Policy Review, Vol. 11, pp. 121–43. Available via the Internet: http://www.ekonomiskaradet.se/Panda_ekonomiska/Data/Documents/sepr2004/DeJong_Lindeboom.pdf.

    • Search Google Scholar
    • Export Citation

  • Drago, R., and M. Wooden, 1992, “The Determinants of Labor Absence: Economic Factors and Workgroup Norms Across Countries,” Industrial and Labor Relations Review, Vol. 45 (July), pp. 764–78.

    • Search Google Scholar
    • Export Citation

  • Drukker, D.M., 2003, “Testing for Serial Correlation in Linear Panel-Data Models,” Stata Journal, Vol. 3 (June), pp. 168–77.

  • Dunn, L.F., and S.A. Youngblood, 1986, “Absenteeism as a Mechanism for Approaching an Optimal Labor Market Equilibrium: An Empirical Study,” Review of Economics and Statistics, Vol. 68 (November), pp. 668–74.

    • Search Google Scholar
    • Export Citation

  • European Foundation for the Improvement of Living and Working Conditions, European Industrial Relations Observatory (EIRO) (Dublin, various issues). Available via the Internet: http://www.eiro.eurofound.eu.int.

    • Search Google Scholar
    • Export Citation

  • Eurostat, New Cronos Database (Luxembourg, Statistical Office of the European Communities).

  • Galí, J., M. Gertler, and J.D. Lopez-Salido, 2005, “Markups, Gaps, and the Welfare Costs of Business Fluctuations,” Economics Working Paper No. 836 (Barcelona, Department of Economics and Business, Universitat Pompeu Fabra). Available via the Internet: http://www.econ.upf.edu/docs/papers/downloads/836.pdf.

    • Search Google Scholar
    • Export Citation

  • Henrekson, M., and M. Persson, 2004, “The Effects on Sick Leave of Changes in the Sickness Insurance System,” Journal of Labor Economics, Vol. 22 (June), pp. 87–114.

    • Search Google Scholar
    • Export Citation

  • Ichino, A., and R.T. Riphahn, 2005, “The Effect of Employment Protection on Worker Effort: A Comparison of Absenteeism During and After Probation,” Journal of the European Economic Association, Vol. 3 (March), pp. 120–43.

    • Search Google Scholar
    • Export Citation

  • Im, K.S., M.H. Pesaran, and Y. Shin, 1997, Testing for Unit Roots in Heterogeneous Panels (Cambridge, United Kingdom, University of Cambridge, Department of Applied Economics, December) Available via the Internet: http://www.econ.cam.ac.uk/faculty/pesaran/lm.pdf.

    • Search Google Scholar
    • Export Citation

  • International Labor Organization (ILO), 2003, Key Indicators of the Labour Markets (KILM) (Geneva, International Labor Organization, 3rd ed.).

    • Search Google Scholar
    • Export Citation

  • International Monetary Fund (IMF), 2003, World Economic Outlook, April 2003: A Survey by the Staff of the International Monetary Fund World Economic and Financial Surveys (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • International Monetary Fund (IMF), 2004, World Economic Outlook, April 2004: A Survey by the Staff of the International Monetary Fund World Economic and Financial Surveys (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Johansson, P., and M. Palme, 1996, “Do Economic Incentives Affect Work Absence? Empirical Evidence Using Swedish Micro Data,” Journal of Public Economics, Vol. 59 (February), pp. 195–218.

    • Search Google Scholar
    • Export Citation

  • Johansson, P., and M. Palme, 2002, “Assessing the Effect of Public Policy on Worker Absenteeism,” Journal of Human Resources, Vol. 37 (Spring), pp. 381–409.

    • Search Google Scholar
    • Export Citation

  • Judson, R.A., and A.L. Owen, 1999, “Estimating Dynamic Panel Data Models: A Guide for Macroeconomists,” Economics Letters, Vol. 65 (October), pp. 9–15.

    • Search Google Scholar
    • Export Citation

  • Kaivanto, K., 1997, “An Alternative Model of Pro-Cyclical Absenteeism,” Economics Letters, Vol. 54 (January), pp. 29–34.

  • Kiviet, J.F., 1995, “On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models,” Journal of Econometrics, Vol. 68 (July), pp. 53–78.

    • Search Google Scholar
    • Export Citation

  • Kocherlakota, N., 2003, “Simplifying Optimal Unemployment Insurance: The Impact of Hidden Savings,” (unpublished; Federal Reserve Bank of Minneapolis).

    • Search Google Scholar
    • Export Citation

  • Larsson, L., 2002, “Sick of Being Unemployed? Interactions Between Unemployment and Sickness Insurance in Sweden,” IFAU Working Paper 2002:6 (Uppsala, Sweden, Institute for Labour Market Policy Evaluation). Available via the Internet: http://www.ifau.se/upload/pdf/se/2002/wp02-06.pdf.

    • Search Google Scholar
    • Export Citation

  • Larsson, L., 2004, “Harmonizing Unemployment and Sickness Insurance: Why (Not)?,” IFAU Working Paper 2004: 8 (Uppsala, Sweden, Institute for Labour Market Policy Evaluation). Available via the Internet: http://www.ifau.se/upload/pdf/se/2004/wp04-08.pdf.

    • Search Google Scholar
    • Export Citation

  • Leigh, J.P., 1985, “The Effects of Unemployment and the Business Cycle on Absenteeism,” Journal of Economics and Business, Vol. 37 (May), pp. 159–70.

    • Search Google Scholar
    • Export Citation

  • Lentz, R., 2007, “Optimal Unemployment Insurance in an Estimated Job Search Model with Savings,” (unpublished; Madison, University of Wisconsin). Available via the Internet: http://www.ssc.wisc.edu/~rlentz/ressources/structui.pdf.

    • Search Google Scholar
    • Export Citation

  • Lentz, R., and T. Traæs, 2005, “Job Search and Savings: Wealth Effects and Duration Dependence,” Journal of Labor Economics, Vol. 23 (July), pp. 467–90.

    • Search Google Scholar
    • Export Citation

  • Levin, A., and C. Lin, 1992, “Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties,” Economics Working Paper 92-23 (La Jolla, California, University of California, San Diego). Available via the Internet: http://www.econ.ucsd.edu/papers/files/ucsd9223.pdf.

    • Search Google Scholar
    • Export Citation

  • Lindbeck, A., 1997, “The Swedish Experiment,” Journal of Economic Literature, Vol. 35 (September), pp. 1273–319.

  • Lusinyan, L., 2005, “Dynamic Panel Data Models in Macroeconomic Applications: The Role of Cross-Sectional Dependence,” presented at the 12th Conference on Panel Data, Copenhagen, June (Ph.D. thesis; Florence, European University Institute).

    • Search Google Scholar
    • Export Citation

  • Mulligan, C., 2002, “A Century of Labor-Leisure Distortions,” NBER Working Paper No. 8774 (Cambridge, Massachusetts, National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Mutual Information System on Social Protection in the EU Member States and the EEA (MISSOC), 2006, “Social Protection in the EU Member States and the European Economic Area (Brussels, European Commission, Directorate-General for Employment and Social Affairs). Available via the Internet: http://europa.eu.int/comm/employment_social/social_protection/missoc_en.htm.

    • Search Google Scholar
    • Export Citation

  • Nickell, S., and L. Nunziata, 2001, Labour Market Institutions Database (unpublished; London, Centre for Economic Performance, London School of Economics).

    • Search Google Scholar
    • Export Citation

  • Nickell, S., and L. Nunziata, W. Ochel, and G. Quintini, 2001, “The Beveridge Curve, Unemployment and Wages in the OECD from the 1960s to the1990s,” CEP Discussion Paper No. 0502 (London, Centre for Economic Performance, London School of Economics). Available via the Internet: http://cep.lse.ac.uk/pubs/download/DP0502.pdf.

    • Search Google Scholar
    • Export Citation

  • Organization for Economic Cooperation and Development (OECD), 2003, The Sources of Economic Growth in OECD Countries (Paris, Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation

  • Organization for Economic Cooperation and Development (OECD) Economic Outlook Database (Paris, Organization for Economic Cooperation and Development) Available via the Internet: http://www.sourceoecd.org.

    • Search Google Scholar
    • Export Citation

  • Organization for Economic Cooperation and Development (OECD), Employment Outlook, Statistical Annex (Paris, various issues). Available via the Internet: http://www.oecd.org/document/37/0,2340,en_2649_34731_31736485_1_1_1_1,00.html.

    • Search Google Scholar
    • Export Citation

  • Organization for Economic Cooperation and Development (OECD) Social Expenditure Database (SOCX) (Paris, Organization for Economic Cooperation and Development) Available via the Internet: http://www.oecd.org/document/2/0,2340,en_2649_34635_31612994_1_1_1_1,00.html.

    • Search Google Scholar
    • Export Citation

  • Palme, M., and I. Svensson, 2003, “Pathways to Retirement and Retirement Incentives in Sweden,” in Alternatives for Welfare Policy, ed. by T. Andersen and P. Molander (Cambridge, United Kingdom, Cambridge University Press).

    • Search Google Scholar
    • Export Citation

  • Prescott, E.C., 2004, “Why Do Americans Work So Much More Than Europeans?” NBER Working Paper No. 10316 (Cambridge, Massachusetts, National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Scruggs, L., 2004, Comparative Welfare Entitlements Data Set (Storrs, Connecticut, University of Connecticut). Available via the Internet: http://sp.uconn.edu/~scruggs/wp.htm.

    • Search Google Scholar
    • Export Citation

  • Shapiro, C., and J.E. Stiglitz, 1984, “Equilibrium Unemployment as a Worker Discipline Device,” American Economic Review, Vol. 74 (June), pp. 433–44.

    • Search Google Scholar
    • Export Citation

  • Skogman Thoursie, P., 2002, “Reporting Sick: Are Sporting Events Contagious?” Research Paper in Economics 2002: 4 (Stockholm, Department of Economics, Stockholm University).

    • Search Google Scholar
    • Export Citation

  • United States Social Security Administration (U.S. SSA) (various issues) Social Security Programs Throughout the World (Washington, United States Social Security Administration), Available via the Internet: http://www.ssa.gov/policy/docs/progdesc/ssptw/.

    • Search Google Scholar
    • Export Citation

  • Werning, I., 2002, “Optimal Unemployment Insurance with Unobservable Savings,” (unpublished; Cambridge, Massachusetts, Massachusetts Institute of Technology). Available via the Internet: http://econ-www.mit.edu/faculty/download_pdf.php?id = 902.

    • Search Google Scholar
    • Export Citation

  • Wooldridge, J.M, 2002, Econometric Analysis of Cross Section and Panel Data (Cambridge, Massachusetts, MIT Press).

*

Lusine Lusinyan is an economist with the IMF Fiscal Affairs Department. Leo Bonato is a senior economist with the IMF Middle East and Central Asia Department. This paper has benefited from comments from Krister Andersson; Marcello Estevão; Robert Flood; seminar participants at the Swedish Ministry of Finance, the Institute for Labour Market Policy Evaluation in Uppsala, and IMF headquarters; and an anonymous referee. The authors wish to thank the Eurostat New Cronos LFS team, Lyle Scruggs, and Xavier Debrun for providing inputs to the data set; Haiyan Shi for excellent research assistance with the data from the U.S. Social Security Administration; and Subhash Thakur for helpful advice.

1

The data on average hours are intended for comparisons of trends over time and are not suitable for comparisons of levels (Organization for Economic Cooperation and Development (OECD), OECD Employment Outlook, Statistical Annex, various issues).

2

See European Foundation for the Improvement of Living and Working Conditions (various issues).

3

Indeed, even though public sickness benefits as a percentage of GDP have generally declined during the past two decades, they remained on average higher than 1 percent of GDP in the Netherlands, Sweden, and Norway over 1990–99 (OECD Social Expenditure Database).

4

In Figure 4, cyclical fluctuations are proxied by the unemployment gap, which is defined as the percentage deviation of the unemployment rate from its linear trend (relations remain largely robust to using a deviation from the quadratic trend).

5

There is a large body of Swedish literature providing empirical evidence of strong moral hazard effects of the insurance system. See, for example, Andrén (2001a, 2001b, and 2003); Johansson and Palme (1996 and 2002); Skogman Thoursie; and Henrekson and Persson (2004). Skogman Thoursie (2002), for example, finds a noticeable increase in men’s sickness absence when popular sports events take place. The interaction of sickness insurance with other elements of the social insurance system, especially unemployment insurance may also produce perverse incentives (Larsson, 2002 and 2004; and Palme and Svensson, 2003).

6

This measure does not include the costs to employers arising from separate provisions negotiated with workers.

7

The gap between agreed to and usual hours arguably reflects the United Kingdom’s long hours and overtime culture and the low coverage of collective bargaining (European Foundation for the Improvement of Living and Working Conditions).

8

See, for example, Allen (1981) and Leigh (1985). Brown and Sessions (1996) provide an extensive survey of the theoretical literature on labor absence.

9

We focus only on work absence; a more general setup could consider also overtime work, such that a∈[—(1—c), c], where a ∈=—(1—c) would be the extreme case of overtime work equal to the total time remaining after the contracted hours of work.

10

The following standard assumptions about theutility function aremade: Ux(.)>0, Ul(.)>0, Uxx(.)<0, Ull(.)<0, Uxl(.) = 0.

11

In this static setting, the paper follows the literature by capturing savings or wealth with the nonlabor incomeand abstracts from a savings decision. Introducing dynamics along with a possibility of self-insurance through savings can be an interesting extension and provide useful insights into such policy issues as optimal sickness insurance. For such extensions in the context of job search models and implications for optimal unemployment benefit policy, see, for example, Werning (2002), Kocherlakota (2003), Lentz (2007), and Lentz and Tranæs (2005). We owe the latter observation to an anonymous referee.

12

Looking at labor-leisure distortions for the U.S. economy over the past century, Mulligan (2002) discusses a number of factors that can drive a wedge between MRS and MPL, including marginal tax rates, transfer payments, labor market regulations, monopoly unions, and the unemployment rate. At the business cycle frequency, Galí, Gertler, and Lόpez-Salido (2005) relate the wedge to price and wage markups.

13

Equation (9) can be obtained by assuming in addition that w(c-βa)-(B+G) > 0 and Paa(.)≥0. It can be shown that Equation (9) simplifies to the solution presented in Prescott (2004), assuming U(x, l)=(1-σ)ln x+σln l and B=0, with σє[0,1] interpreted as the value of the leisure or sickness index.

14

The latter condition requires some strict assumptions on the probability function to be inversely dependent on a (only or separably from v), and, given that P is a probability and a<1, on the constant term in the solution of the differential equation. Note also that D has the same sign as N and that given our assumptions on the function P, Na<0.

15

Given a lack of consistent cross-country data on wealth or nonlabor income, the latter has not been included in the empirical analysis.

16

Employees are grouped into two main subgroups: those who worked at least one hour during the reference week, and those who had a job, but did not work at all during the reference week. For the first group there are 13 reasons provided for absence—defined as a positive difference between usual and actual hours of work—and nine for the second group. We refer to sickness absence as absence because of a worker’s own illness, injury, or temporary disability. Sickness absence of those in the first group is defined as “short-term” and that of those in the second group is defined as “long-term.” Absence spells tend to be considerably longer than one week in all countries except the United Kingdom and Iceland, but long-term sickness is of particular concern because it is likely to change into disability status. Palme and Svensson (2003) show that in Sweden this has become one of the most common ways to exit the labor force before the statutory retirement age.

17

The LSDV estimator with the Kiviet’;s (1995) correction outperforms all other estimators in small samples, but an implementation of this technique for unbalanced panels had not been derived. Recently, Bruno (2004) proposed an extension of the Kiviet’s correction to unbalanced panels with strictly exogenous covariates, using a bootstrap approach to estimate the variance-covariance matrix.

18

The dependent variable has been tested for nonstationarity using Levin and Lin (1992) and Im, Pesaran, and Shin (1997) panel unit root tests. The results, not reported here, indicate that the null of nonstationarity can be rejected when a trend is included in the specification.

19

Moreover, for the Arellano-Bond estimator, Bun and Kiviet (2006) show that reducing the total number of instruments by a factor T reduces the bias in a finite sample by a factor T.

20

Because our dependent variable is the sickness absence rate among full-time employees in a given year, independent of sick leave status, the employees would be part of the labor force that year, with their life expectancy at birth also given.

21

In particular, based on the estimates of the contemporaneous impact coefficients, a 1 percentage point increase in the average participation rate would increase the average absence rate from 2.75 to 2.8 percent; a one-year increase in life expectancy is estimated to lower the average absence rate to 2.65 percent (estimates of the elasticity at the mean of about 1.25 and—2.8 are used for the participation rate and life expectancy, respectively).

22

The impact of privately financed insurance, in the model, has been shown to depend on the model assumptions, particularly concerning the sensitivity of employment continuation with respect to absence behavior. In the robustness analysis, we look into two subsamples with more and less stringent employment protection and with a higher and lower level of unionization, factors that, to a large extent, could determine the degree of such sensitivity. We find support for the model’s predictions showing that more private financing of sickness insurance will have a much stronger negative impact on sickness absence in the subsample corresponding to stricter labor regulations.

23

In general, however, the direct impact of labor market institutions on absence behavior can be ambiguous, as discussed in the section on the robustness analysis. Alternative measures for employment protection, such as the index measuring the strictness of employment protection and the degree of wage bargaining coordination, are also discussed.

24

Allowing observations to be correlated within countries but independent across countries lowers the statistical significance of life expectancy while making labor force participation slightly more significant.

25

The high and persistent level of employment protection in Europe (IMF, 2003 and 2004) would support a weak impact of absence behavior on changes in working arrangements. We owe this observation to an anonymous referee.

26

It is common in the literature on labor market institutions to assume that institutional variables are exogenous (IMF, 2003).

27

In addition, other estimates (not reported here) indicate that the impact of working-time arrangements is more significant and robust to changes in the country sample and the type of absence if: (i) instead of using robust variance estimators in the static models, the disturbance term is assumed to be of a first-order autoregressive form; and (ii) richer dynamics, particularly in terms of higher autoregressive lags, are assumed in the dynamic models.

28

These results apply to 13 out of 18 countries in our sample, for which data on sickness and unemployment insurance schemes are available (see Figure 5).

29

In Appendix Table A.8, column 2, for example, the t-value of sickness benefits would drop to 1.36 from the reported 1.87 when life expectancy is included, and to 1.29 when Sweden is excluded, partly owing to a smaller coefficient estimate of 0.01. Indeed, a significant and large positive interaction term of Sweden’s fixed effect with sickness benefits indicates that the impact for Sweden is substantially stronger than for the cross-country average (not reported here).

30

See the notes in Table A.8 for more details concerning the choice of subsamples. Given the resulting small size of the subsamples when partitioning the data, we do not report the results from dynamic panel data estimations.

31

Measurement issues, including the degree of enforcement of labor market regulations, arise when using such institutional indicators; however, they are considered to reasonably proxy the most relevant institutional features (IMF, 2003).

32

In the literature on wage determination and unemployment (for example, Nickell and Nunziata, 2001; Nickell and others, 2001; and IMF, 2003), coordination refers to mechanisms on both the unions’ and employers’ sides—such as centralized bargaining (as in Austria) or the presence of institutions assisting in bargaining (as in Germany)—whereby the aggregate employment implications of wage determination are taken into account when wage bargains are struck.

33

Ichino and Riphahn (2005) provide evidence of a significant positive impact of employment protection on absence rates for Italy.

34

The Swedish example shows that the interaction of sickness insurance with unemployment insurance creates a perverse incentive for the unemployed to be listed as sick. By harmonizing the replacement rates between the two systems in 2003, the government has largely reduced this incentive. A review of the link between sickness insurance and disability pensions and their role in promoting early retirement is also desirable.

Same Series

Other IMF Content

Other Publishers

  • Collapse
  • Expand
Cover IMF Staff Papers
IMF Staff Papers, Volume 54, No. 3
Author:
International Monetary Fund. Research Dept.
Volume/Issue:
Volume 2007: Issue 001
Publisher:
International Monetary Fund
ISBN:
9781589066519
ISSN:
1020-7635
Pages:
211
DOI:
https://doi.org/10.5089/9781589066519.024
Figure 1.

Labor Force Participation Rate and Average Hours Worked Annually per Employee
Figure 2.

Average Sickness Absence, 1995–2003

(As a percentage of employment)
Figure 3.

Sickness Absence

(Employees absent due to sickness as a percentage of total employed)
Figure 4.

Cyclicality of Sickness Absence
Figure 5.

Changes in Sickness Insurance System, 1983–2002
Figure 6.

Sickness Cash Benefits Paid by Employer
Figure 7.

Working-Time Arrangements